This paper focuses on the problem of script identification in scene text images. Facing this problem with state of the art CNN classifiers is not straightforward, as they fail to address a key characteristic of scene text instances: their extremely variable aspect ratio. Instead of resizing input images to a fixed aspect ratio as in the typical use of holistic CNN classifiers, we propose here a patch-based classification framework in order to preserve discriminative parts of the image that are characteristic of its class. We describe a novel method based on the use of ensembles of conjoined networks to jointly learn discriminative stroke-parts representations and their relative importance in a patch-based classification scheme. Our experiments with this learning procedure demonstrate state-of-the-art results in two public script identification datasets. In addition, we propose a new public benchmark dataset for the evaluation of multi-lingual scene text end-to-end reading systems. Experiments done in this dataset demonstrate the key role of script identification in a complete end-to-end system that combines our script identification method with a previously published text detector and an off-the-shelf OCR engine.
Class imbalance is one of the challenging problems for machine learning in many real-world applications, such as coal and gas burst accident monitoring: the burst premonition data is extreme smaller than the normal data, however, which is the highlight we truly focus on. Cost-sensitive adjustment approach is a typical algorithm-level method resisting the data set imbalance. For SVMs classifier, which is modified to incorporate varying penalty parameter(C) for each of considered groups of examples. However, the C value is determined empirically, or is calculated according to the evaluation metric, which need to be computed iteratively and time consuming. This paper presents a novel cost-sensitive SVM method whose penalty parameter C optimized on the basis of cluster probability density function(PDF) and the cluster PDF is estimated only according to similarity matrix and some predefined hyper-parameters. Experimental results on various standard benchmark data sets and real-world data with different ratios of imbalance show that the proposed method is effective in comparison with commonly used cost-sensitive techniques.
We revisit two popular convolutional neural networks (CNN) in person re-identification (re-ID), i.e, verification and classification models. The two models have their respective advantages and limitations due to different loss functions. In this paper, we shed light on how to combine the two models to learn more discriminative pedestrian descriptors. Specifically, we propose a new siamese network that simultaneously computes identification loss and verification loss. Given a pair of training images, the network predicts the identities of the two images and whether they belong to the same identity. Our network learns a discriminative embedding and a similarity measurement at the same time, thus making full usage of the annotations. Albeit simple, the learned embedding improves the state-of-the-art performance on two public person re-ID benchmarks. Further, we show our architecture can also be applied in image retrieval.
In this work, we propose a novel information theoretic framework for dictionary learning (DL) and sparse coding (SC) on a statistical manifold (the manifold of probability distributions). Unlike the traditional DL and SC framework, our new formulation {\it does not explicitly incorporate any sparsity inducing norm in the cost function but yet yields SCs}. Moreover, we extend this framework to the manifold of symmetric positive definite matrices, $\mathcal{P}_n$. Our algorithm approximates the data points, which are probability distributions, by the weighted Kullback-Leibeler center (KL-center) of the dictionary atoms. The KL-center is the minimizer of the maximum KL-divergence between the unknown center and members of the set whose center is being sought. Further, {\it we proved that this KL-center is a sparse combination of the dictionary atoms}. Since, the data reside on a statistical manifold, the data fidelity term can not be as simple as in the case of the vector-space data. We therefore employ the geodesic distance between the data and a sparse approximation of the data element. This cost function is minimized using an acceleterated gradient descent algorithm. An extensive set of experimental results show the effectiveness of our proposed framework. We present several experiments involving a variety of classification problems in Computer Vision applications. Further, we demonstrate the performance of our algorithm by comparing it to several state-of-the-art methods both in terms of classification accuracy and sparsity.
Regularized discriminant analysis (RDA), proposed by Friedman (1989), is a widely popular classifier that lacks interpretability and is impractical for high-dimensional data sets. Here, we present an interpretable and computationally efficient classifier called high-dimensional RDA (HDRDA), designed for the small-sample, high-dimensional setting. For HDRDA, we show that each training observation, regardless of class, contributes to the class covariance matrix, resulting in an interpretable estimator that borrows from the pooled sample covariance matrix. Moreover, we show that HDRDA is equivalent to a classifier in a reduced-feature space with dimension approximately equal to the training sample size. As a result, the matrix operations employed by HDRDA are computationally linear in the number of features, making the classifier well-suited for high-dimensional classification in practice. We demonstrate that HDRDA is often superior to several sparse and regularized classifiers in terms of classification accuracy with three artificial and six real high-dimensional data sets. Also, timing comparisons between our HDRDA implementation in the sparsediscrim R package and the standard RDA formulation in the klaR R package demonstrate that as the number of features increases, the computational runtime of HDRDA is drastically smaller than that of RDA.
Over the last decade, proliferation of various online platforms and their increasing adoption by billions of users have heightened the privacy risk of a user enormously. In fact, security researchers have shown that sparse microdata containing information about online activities of a user although anonymous, can still be used to disclose the identity of the user by cross-referencing the data with other data sources. To preserve the privacy of a user, in existing works several methods (k-anonymity, l-diversity, differential privacy) are proposed that ensure a dataset which is meant to share or publish bears small identity disclosure risk. However, the majority of these methods modify the data in isolation, without considering their utility in subsequent knowledge discovery tasks, which makes these datasets less informative. In this work, we consider labeled data that are generally used for classification, and propose two methods for feature selection considering two goals: first, on the reduced feature set the data has small disclosure risk, and second, the utility of the data is preserved for performing a classification task. Experimental results on various real-world datasets show that the method is effective and useful in practice.
Connectionist temporal classification (CTC) based supervised sequence training of recurrent neural networks (RNNs) has shown great success in many machine learning areas including end-to-end speech and handwritten character recognition. For the CTC training, however, it is required to unroll (or unfold) the RNN by the length of an input sequence. This unrolling requires a lot of memory and hinders a small footprint implementation of online learning or adaptation. Furthermore, the length of training sequences is usually not uniform, which makes parallel training with multiple sequences inefficient on shared memory models such as graphics processing units (GPUs). In this work, we introduce an expectation-maximization (EM) based online CTC algorithm that enables unidirectional RNNs to learn sequences that are longer than the amount of unrolling. The RNNs can also be trained to process an infinitely long input sequence without pre-segmentation or external reset. Moreover, the proposed approach allows efficient parallel training on GPUs. For evaluation, phoneme recognition and end-to-end speech recognition examples are presented on the TIMIT and Wall Street Journal (WSJ) corpora, respectively. Our online model achieves 20.7% phoneme error rate (PER) on the very long input sequence that is generated by concatenating all 192 utterances in the TIMIT core test set. On WSJ, a network can be trained with only 64 times of unrolling while sacrificing 4.5% relative word error rate (WER).
Statistical Relational Learning (SRL) methods have shown that classification accuracy can be improved by integrating relations between samples. Techniques such as iterative classification or relaxation labeling achieve this by propagating information between related samples during the inference process. When only a few samples are labeled and connections between samples are sparse, collective inference methods have shown large improvements over standard feature-based ML methods. However, in contrast to feature based ML, collective inference methods require complex inference procedures and often depend on the strong assumption of label consistency among related samples. In this paper, we introduce new relational features for standard ML methods by extracting information from direct and indirect relations. We show empirically on three standard benchmark datasets that our relational features yield results comparable to collective inference methods. Finally we show that our proposal outperforms these methods when additional information is available.
We adapted the join-training scheme of Faster RCNN framework from Caffe to TensorFlow as a baseline implementation for object detection. Our code is made publicly available. This report documents the simplifications made to the original pipeline, with justifications from ablation analysis on both PASCAL VOC 2007 and COCO 2014. We further investigated the role of non-maximal suppression (NMS) in selecting regions-of-interest (RoIs) for region classification, and found that a biased sampling toward small regions helps performance and can achieve on-par mAP to NMS-based sampling when converged sufficiently.
This paper presents a novel model for multimodal learning based on gated neural networks. The Gated Multimodal Unit (GMU) model is intended to be used as an internal unit in a neural network architecture whose purpose is to find an intermediate representation based on a combination of data from different modalities. The GMU learns to decide how modalities influence the activation of the unit using multiplicative gates. It was evaluated on a multilabel scenario for genre classification of movies using the plot and the poster. The GMU improved the macro f-score performance of single-modality approaches and outperformed other fusion strategies, including mixture of experts models. Along with this work, the MM-IMDb dataset is released which, to the best of our knowledge, is the largest publicly available multimodal dataset for genre prediction on movies.
This paper presents a supervised Aspect Based Sentiment Analysis (ABSA) system. Our aim is to develop a modular platform which allows to easily conduct experiments by replacing the modules or adding new features. We obtain the best result in the Opinion Target Extraction (OTE) task (slot 2) using an off-the-shelf sequence labeler. The target polarity classification (slot 3) is addressed by means of a multiclass SVM algorithm which includes lexical based features such as the polarity values obtained from domain and open polarity lexicons. The system obtains accuracies of 0.70 and 0.73 for the restaurant and laptop domain respectively, and performs second best in the out-of-domain hotel, achieving an accuracy of 0.80.
Deep Neural Networks, and specifically fully-connected convolutional neural networks are achieving remarkable results across a wide variety of domains. They have been trained to achieve state-of-the-art performance when applied to problems such as speech recognition, image classification, natural language processing and bioinformatics. Most of these deep learning models when applied to classification employ the softmax activation function for prediction and aim to minimize cross-entropy loss. In this paper, we have proposed a supervised model for dominant category prediction to improve search recall across all eBay classifieds platforms. The dominant category label for each query in the last 90 days is first calculated by summing the total number of collaborative clicks among all categories. The category having the highest number of collaborative clicks for the given query will be considered its dominant category. Second, each query is transformed to a numeric vector by mapping each unique word in the query document to a unique integer value; all padded to equal length based on the maximum document length within the pre-defined vocabulary size. A fully-connected deep convolutional neural network (CNN) is then applied for classification. The proposed model achieves very high classification accuracy compared to other state-of-the-art machine learning techniques.
Point patterns are sets or multi-sets of unordered elements that can be found in numerous data sources. However, in data analysis tasks such as classification and novelty detection, appropriate statistical models for point pattern data have not received much attention. This paper proposes the modelling of point pattern data via random finite sets (RFS). In particular, we propose appropriate likelihood functions, and a maximum likelihood estimator for learning a tractable family of RFS models. In novelty detection, we propose novel ranking functions based on RFS models, which substantially improve performance.
Morphology in unbalanced languages remains a big challenge in the context of machine translation. In this paper, we propose to de-couple machine translation from morphology generation in order to better deal with the problem. We investigate the morphology simplification with a reasonable trade-off between expected gain and generation complexity. For the Chinese-Spanish task, optimum morphological simplification is in gender and number. For this purpose, we design a new classification architecture which, compared to other standard machine learning techniques, obtains the best results. This proposed neural-based architecture consists of several layers: an embedding, a convolutional followed by a recurrent neural network and, finally, ends with sigmoid and softmax layers. We obtain classification results over 98% accuracy in gender classification, over 93% in number classification, and an overall translation improvement of 0.7 METEOR.
This paper presents a stochastic logic time delay reservoir design. The reservoir is analyzed using a number of metrics, such as kernel quality, generalization rank, performance on simple benchmarks, and is also compared to a deterministic design. A novel re-seeding method is introduced to reduce the adverse effects of stochastic noise, which may also be implemented in other stochastic logic reservoir computing designs, such as echo state networks. Benchmark results indicate that the proposed design performs well on noise-tolerant classification problems, but more work needs to be done to improve the stochastic logic time delay reservoir's robustness for regression problems.
When integrating computational tools such as automatic segmentation into clinical practice, it is of utmost importance to be able to assess the level of accuracy on new data, and in particular, to detect when an automatic method fails. However, this is difficult to achieve due to absence of ground truth. Segmentation accuracy on clinical data might be different from what is found through cross-validation because validation data is often used during incremental method development, which can lead to overfitting and unrealistic performance expectations. Before deployment, performance is quantified using different metrics, for which the predicted segmentation is compared to a reference segmentation, often obtained manually by an expert. But little is known about the real performance after deployment when a reference is unavailable. In this paper, we introduce the concept of reverse classification accuracy (RCA) as a framework for predicting the performance of a segmentation method on new data. In RCA we take the predicted segmentation from a new image to train a reverse classifier which is evaluated on a set of reference images with available ground truth. The hypothesis is that if the predicted segmentation is of good quality, then the reverse classifier will perform well on at least some of the reference images. We validate our approach on multi-organ segmentation with different classifiers and segmentation methods. Our results indicate that it is indeed possible to predict the quality of individual segmentations, in the absence of ground truth. Thus, RCA is ideal for integration into automatic processing pipelines in clinical routine and as part of large-scale image analysis studies.
Recently, machine learning methods have provided a broad spectrum of original and efficient algorithms based on Deep Neural Networks (DNN) to automatically predict an outcome with respect to a sequence of inputs. Recurrent hidden cells allow these DNN-based models to manage long-term dependencies such as Recurrent Neural Networks (RNN) and Long Short-Term Memory (LSTM). Nevertheless, these RNNs process a single input stream in one (LSTM) or two (Bidirectional LSTM) directions. But most of the information available nowadays is from multistreams or multimedia documents, and require RNNs to process these information synchronously during the training. This paper presents an original LSTM-based architecture, named Parallel LSTM (PLSTM), that carries out multiple parallel synchronized input sequences in order to predict a common output. The proposed PLSTM method could be used for parallel sequence classification purposes. The PLSTM approach is evaluated on an automatic telecast genre sequences classification task and compared with different state-of-the-art architectures. Results show that the proposed PLSTM method outperforms the baseline n-gram models as well as the state-of-the-art LSTM approach.
This paper introduces a Deep Scattering network that utilizes Dual-Tree complex wavelets to extract translation invariant representations from an input signal. The computationally efficient Dual-Tree wavelets decompose the input signal into densely spaced representations over scales. Translation invariance is introduced in the representations by applying a non-linearity over a region followed by averaging. The discriminatory information in the densely spaced, locally smooth, signal representations aids the learning of the classifier. The proposed network is shown to outperform Mallat's ScatterNet on four datasets with different modalities on classification accuracy.
We introduce a ScatterNet that uses a parametric log transformation with Dual-Tree complex wavelets to extract translation invariant representations from a multi-resolution image. The parametric transformation aids the OLS pruning algorithm by converting the skewed distributions into relatively mean-symmetric distributions while the Dual-Tree wavelets improve the computational efficiency of the network. The proposed network is shown to outperform Mallat's ScatterNet on two image datasets, both for classification accuracy and computational efficiency. The advantages of the proposed network over other supervised and some unsupervised methods are also presented using experiments performed on different training dataset sizes.
Forward-looking ground-penetrating radar (FLGPR) has recently been investigated as a remote sensing modality for buried target detection (e.g., landmines). In this context, raw FLGPR data is beamformed into images and then computerized algorithms are applied to automatically detect subsurface buried targets. Most existing algorithms are supervised, meaning they are trained to discriminate between labeled target and non-target imagery, usually based on features extracted from the imagery. A large number of features have been proposed for this purpose, however thus far it is unclear which are the most effective. The first goal of this work is to provide a comprehensive comparison of detection performance using existing features on a large collection of FLGPR data. Fusion of the decisions resulting from processing each feature is also considered. The second goal of this work is to investigate two modern feature learning approaches from the object recognition literature: the bag-of-visual-words and the Fisher vector for FLGPR processing. The results indicate that the new feature learning approaches outperform existing methods. Results also show that fusion between existing features and new features yields little additional performance improvements.